My dependent variable is Choice
, 0 for non-issuers, 1 for seasoned equity issuers, 2 for convertible issuers and 3 for bond issuers.
I use the following command to run a multinomial logit regression:
mlogit Choice CR $ConVar, base(0)
To test for IIA assumption, I use the following command:
mlogtest, haus
The output is below:
Does this result indicate that IIA is violated?
Then I test the IIA assumption for another multinomial logit regression, in which the dependent variable is Security
(0 for seasoned equity issuers, 1 for convertible issuers, and 2 for bond issuers).
The Stata commands are below:
mlogit Security CR $ConVar, base(0)
mlogtest, haus
Output of IIA test is shown below:
Again, does this mean that the IIA assumption is violated?
Could anyone interpret these two results for me? And I am not sure why some categories of dependent variables have negative chi2.
Best Answer
Significant values of chi2 indicate that the IIA assumption has been violated. You can look at the p-values in the 4th column to gauge that.
Some of your test statistics are negative, which is fairly common. Hausman and McFadden (1984, p. 1226) argue that a negative result is evidence that IIA has not been violated.
This logic would suggest that the first one passes, and the second fails.
A good, accessible book with lots of Stata examples and much intuition on this is Long & Freese's Regression Models for Categorical Dependent Variables Using Stata.
The (partially gated) paper mentioned above is: Hausman, Jerry, and Daniel McFadden. "Specification Tests for the Multinomial Logit Model." Econometrica 52, no. 5 (1984): 1219-240.